Modal resonator with hundreds of partials
-
- KVRian
- Topic Starter
- 513 posts since 3 Sep, 2009 from Poland
I've been wondering about this since a longer time. How do some modal synths like e.g. NI Spark handle hundreds of band pass filters in polyphony, while keeping low cpu? Time domain approach , like computing SVFs using SIMD seems not enough. Do they use FFT/circular convolution?
giq
-
- KVRian
- 833 posts since 21 Feb, 2006 from FI
Not an answer but, not long ago I calculated quite simple Fourier Series task in a two for-loops and getting answer took over one hour ... someone guided to use matrix algebra or dot-product and sums which of the latter I implemented and now the same task takes only less than 0.1 sec to finish ... .
- KVRAF
- 7890 posts since 12 Feb, 2006 from Helsinki, Finland
Your question might be a bit too vague, since there's probably not a "one size fits all" answer here. If your filters are static, then computing the IR once and using regular fast convolution is likely ideal, but a couple of hundred SVFs in parallel isn't that much work for ALU either and the performance is likely going to be dominated by the amount of data you need to move around.
ps. No idea how much CPU something like NI Spark uses (or what it even does), so "low CPU" might mean anything as far as I'm concerned.
ps. No idea how much CPU something like NI Spark uses (or what it even does), so "low CPU" might mean anything as far as I'm concerned.